1. Field of the Invention
The present invention relates to an image processing apparatus, an image processing method, a defect detection method, a semiconductor device manufacturing method, and a program.
2. Related Background Art
A conventional art and its problems are described taking as an example a scanning electron microscope (SEM) image of a micropattern formed in a semiconductor manufacturing process.
Dimensional measurement using a CD (critical dimension)-SEM is conducted to evaluate a micropattern of a semiconductor. The CDSEM acquires an SEM image in which a measurement target pattern is observed from above, and the dimensions of parts of the pattern are measured by an apparatus on the basis of the SEM image. Here, the acquired SEM image may be stored in a storage medium such as an HDD connected to an EWS for the purpose of subsequent remeasurement and reobservation. When the image is measured, it is preferable to use a pixel base format such as device independent bitmap (BWP) as a format of the image to be saved, in order to prevent the deterioration of measurement accuracy at the time of measurement. The BMP has been widely spread as a standard image format for Windows (registered trademark) independent of the model of a PC, but has a disadvantage that the file size of an image is increased. For example, the file size runs up to 16 BM in the case of an 8-bit gray scale image having a size of 4096×4096, which wastes storage medium resources for data storage such as the HDD and CPU resources of the PC for executing image processing. There are also problems of an increased load on a network and longer time required for transfer when the image data is transferred via the network.
An image file is compressed by a run length encoding (RLE) method which carries out lossless compression to reduce the size of the file when the image file is stored on, for example, the HDD, and the image file is decompressed when the image therein is used, such that it is possible to save the storage medium resources such as the HDD. However, in the case of a natural image such as the SEM image, effects of the reduction of the image size by the compression are generally small, and new CPU resources and processing time are required for the compression and decompression.
Furthermore, when the image is transferred through a network such as the Internet, it is possible to use joint photographic experts group (JPEG) which is a normally used compression image format with high compression efficiency. The JPEG is an image format developed for the compression and saving of the natural images. The JPEG provides lossless compression, but has an advantage of being capable of reducing the file sizes of some images several ten times. However, while the JPEG allows the designation of a compression ratio at the time of saving the image, the JPEG has disadvantages that high-frequency components are lost and the degree of deterioration becomes higher in information of the edge of the image when the compression ratio is increased and that the measurement accuracy decreases at the time of measurement. Thus, the JPEG image is not suitable to the image format for measurement.
In addition to the JPEG, there has also been proposed a method called vector quantization (VQ) compression as a method suitable to the compression of a line image such as an illustration rather than the natural image. This method is characterized by the use of what is called a code book which is prepared in the following manner: an image is divided into small image blocks, each of the divided image blocks is regarded as one point in a multidimensional space, vector quantization is carried out to reduce the kinds of image blocks, and identification code numbers are assigned to the quantized image blocks.
However, the weakness of this method is the appearance of the seams of the blocks in a reproduced image. Although human eyes can be deceived, this method still remains unsuitable to the measurement as in the case of the PEG.
In measuring the image data via the network, means for reducing the load on the network is especially important when the target of the measurement is an image with a large size. FIG. 8 is a table showing the relation between circuit speeds and transfer times for each image size. It can be understood from FIG. 8 that about three seconds are required to transfer an image of 4096×4096 even by an FTTH circuit of 40 Mbps and that the speed of measurement processing is limited when a large volume of images are measured. Moreover, images targeted for measurement are generally images of products in mass production factories, and when these images as company secrets go outside through the Internet connection, a serious problem is caused in terms of security. Measurements can be taken for this problem by, for example, encrypting the images, but there is another problem of the requirement of heavy CPU resources and processing time for the encoding and decoding for the encryption.